Could AI take over the repetitive work
without needing the same explanation every time?
I run Itaewon Guide, a local magazine on Instagram. Every week I look for places to feature, check the material, write the copy, and make the designs and videos. The subjects changed, but moving material, adjusting layouts, and explaining revisions kept repeating.
A thought occurred to me. If I defined the standards to keep and the order of work, could AI carry the repetitive tasks forward without a fresh explanation and a separate request for every step? I wanted to do more than finish this one post. I wanted to build a system gradually, so the next edition could follow the same standards.
So I connected the steps from reading existing material to planning, Figma editing, video covers, and captions. This is not yet a system that creates and publishes on a set day by itself. This account explores what I could delegate and what standards needed to come first as I made requests and reviewed the results.
We need new stories each week.
But do we also need to repeat
the same explanations and tasks?
This case is for solo content creators and small teams. To try it in your own work, you can start with the handoff template and first steps.
First, I separated what changes
from the standards that stay
I had Codex review the previous editor's material, original Illustrator files, production standards, and account analysis on my computer. Before deciding what to make next, I checked what needed to stay consistent.
The magazine's voice, fonts and colors, logo, card dimensions, and fact-checking principles.
The place and topic, title, photos, and newly verified visitor information.
The latest file, copy and layouts I have approved, and anything still needing revision.
I was not trying to freeze everything. The design and voice could carry over, while this week's subject still needed a fresh decision. Separating those roles made it clearer what I should record to avoid repeating explanations.
Putting everything in a folder does not make AI remember it automatically. At the start, specify the standards and latest files it must read. At the end, record what is approved and what still needs work.
Open the prompt for reading existing context
An example adapted from the actual work requests. Replace the files, account, and edition details with your own.
Project: [channel or content name] Material to read: [accessible project folder or document link] Latest file to continue from: [the file I specify] Previous work to use as a benchmark: [one good example] First read the production rules, edition drafts, and current progress in this material. Separate and summarize: 1. Brand standards to keep consistent 2. Requirements specific to this edition 3. Conflicts between old instructions and recent revisions 4. Unverified points that need my decision Prioritize the latest file I specify and the revisions I have approved. Do not revert to discarded drafts. If you cannot access the material or cannot identify the latest version, tell me rather than guessing. Do not edit files yet. First summarize the standards for the work.
Tools and prerequisites for this project
Codex handled material reading and work requests, while Figma held the editable designs. I organized images, videos, drafts, and working notes in local folders for each edition.
The connections and permissions to access files and edit Figma, along with the required fonts, must be ready first. A prompt alone does not connect every app. This article is not an installation guide; it is a case study of setting standards and continuing production in a prepared environment.
A local folder is a place to store material, not a claim that AI runs without the internet. Keep private client material and personal information separate from the files needed for the task.
Before choosing a place,
I thought about the reader's situation
If the brief is only 'introduce a new place,' titles and photos can become interchangeable. For the Say Touche edition, I started with a situation: where could friends meet in Haebangchon? The story began with meeting outside a distinctive, memorable building, then exploring what was inside.
Introduce a newly opened place.
Introduce a place that answers:
'Where should we meet?'
I gave the two weekly editions different roles: Tuesday for discovering a place worth noticing now, and Friday for saving somewhere for the next outing. Even when AI suggested candidates, I needed to know whose decision the content was meant to help before I could choose a subject and title.
'This could become a new landmark' is an interpretation; 'this is already the most popular place' is a different claim. I separated facts verified through official sources from my own reading, and chose titles the actual photos and story could support. I wanted to automate organizing information and drafting, not accelerate unverified claims.
Open the prompt for turning research into a card-by-card plan
An example adapted from the actual work requests. Replace the files, account, and edition details with your own.
Channel: [channel name and voice to preserve] Reader's situation: [who is facing which problem or decision] This edition's subject: [what to feature and the angle] Planned publication date: [date] Reference material: [official sources and approved planning notes] Create a card-by-card content plan from these criteria. If new information is needed, check official sources first. Leave anything you cannot verify marked as unconfirmed. Do not change the approved subject just because another topic is trending. Separate official information from editorial opinion. Record the publication date, event dates, and verification date separately. Do not turn the opinion 'worth noticing' into the claim 'most popular.' Present the result in this order: reader's situation; why feature this now; three title options and the reasoning; each card's role and short copy; required photos; sources and unverified details.
Looking identical mattered less
than being easy to edit next time
I needed more than my familiar tool:
I needed a setup someone else could continue using.
As a designer, I still find Illustrator familiar and comfortable. But the next person using this template might not be a designer. A text edit or photo swap that feels simple to me could be difficult for someone unfamiliar with the tool.
ready for the team and AI to continue.
places to notice
places to notice
- 01 / PrepareApprove copy and photosMaterial for this edition
- 02 / AI editsUpdate the templateText / Photos / Specified positions
- 03 / Human reviewReview, revise, approveLine breaks / Crops / Layout
An illustrative workflow diagram, not a screenshot of either app. AI editing requires a separate integration and appropriate permissions.
Cost mattered too. Someone without an Illustrator subscription would need to pay for access and learn the program to continue in the same tool. Figma lets people open a file in a browser and work together on the same canvas, with a free plan for basic work. I wanted to lower the barrier for the next person to take part.
So I moved toward a setup where I define the design standards, but someone else can replace the weekly copy and photos. The aim was not to make everyone design from scratch. It was to let people change the necessary parts within an agreed template and review them together.
The free plan has limits, including file counts and sharing scope. Check the plan and permissions needed for team editing and AI integration separately. Figma's free-plan guide · Browser-based collaboration
At first I transferred the Illustrator original as an SVG vector file to preserve its appearance. But in this transfer, the lettering remained shapes, making even a sentence difficult to edit. Reproducing the look alone did not make the next edition easier to produce.
So I separated the reference file from the working template. I rebuilt the production headings and body copy so the text could be selected and edited directly. Photos remained replaceable, while the logo and base layout stayed in place.
The appearance transferred, but lettering converted to shapes was difficult to revise directly.
I made the text and photos separately replaceable in the cover, content, and closing templates.
I inserted actual copy and compared fonts, line breaks, photo crops, and clipped text with the original. This was not a reconstruction of every reference card.
One standard became clear through this process. The file must not only look right now; it must let us replace just the necessary parts for the next edition. Before delegating repeat production, I needed an editable structure.
I went beyond AI-written copy
and connected it to actual design edits.
I was already using AI to research, organize writing, and plan content. What I wanted to try next was carrying that copy into the existing design and replacing the photos too. If AI wrote the draft but I still had to transfer everything into the design myself each time, only part of production would become easier.
I prepared a Figma template with independently editable text and photos, then connected AI so it could actually edit that file. With approved copy, selected photos, and specified positions, I could ask it to update the existing template. I checked the resulting line breaks, crops, and layout, then revised what needed attention.
Not asking for a new design every time,
but updating the necessary parts within a design we had already agreed on.
That was closer to the kind of automation I wanted to connect.
Could this work the same way in ChatGPT or Claude?
The important distinction is the connected capability, not the AI's name. Reading Figma material or creating FigJam diagrams is different from editing text and photos in Figma Design. Your environment must support design editing and have permission to edit the file. This case was carried out in Codex; it does not mean I completed the same workflow in Claude.
In Figma's official guidance checked on October 13, 2026, MCP canvas-writing features are available to Full and Dev seats on paid plans, with Dev seats limited to read-only access outside drafts. Check the AI service's terms of access separately as well. Figma's official integration and editing requirements
Open the prompt for an editable design transfer
An example adapted from the actual work requests. Replace the files, account, and edition details with your own.
Source: [Illustrator file] Working document: [Figma document link] Transfer scope: [the cover, content, or closing templates needed] First check whether you can access the source and edit the specified Figma file. If a connection or permission is missing, explain what is needed; do not claim the work is done. Once ready, transfer the design into a reusable editable template. Preserve the source file. First inspect the artboard size, fonts and weights, heading and body positions, accent color, logo, and closing layout. Transfer photos, vector graphics, and text separately. Keep headings and body copy as selectable, editable text rather than shapes. Check whether the original fonts are available. Do not silently substitute missing fonts or write into another account's document. Build the requested templates first, then compare them side by side with the original. Pay particular attention to the logo, small icons, line breaks, photo crops, and clipped text. Distinguish editable text and photos from shapes retained only as references.
I added the movement of a place
to a still cover
For the Say Touche edition, I checked the scene and proportions of footage showing the building's exterior. I placed the Itaewon Guide title and logo over it. The background moves, while the text the viewer needs to read stays in place.
I placed only the title and logo
over the original footage.
I composited the text and logo exported from Figma without a background over the original footage. After changing the title, I asked for a new version using the latest layers and for checks at several points in the video.
The dark checkerboard on the left indicates transparent areas.
Recompositing and exporting at the same dimensions after a title change was a task I could repeatedly ask AI to handle. I still checked which scene to use and where the text should sit. This added the magazine's design to existing footage; it did not generate new footage of the place.
The working environment at the time limited video insertion, so I kept the still cover in Figma and produced the MP4 separately. This does not mean Figma has no video features. Figma's official video feature guide
Video and photo source: Official Say Touche @say_touche video
Cards catch the eye;
captions add a reason to visit
Repeating the card copy in the caption gives readers the same story twice. In the caption, I started with the situation in which the place might be useful, then connected its features and verified details that did not fit on the cards.
Example caption opening'Where should we meet?'
For your next Haebangchon outing, suggest meeting outside Say Touche.Greenery wraps the exterior, and long horizontal windows run across it. Its distinctive facade stands out among neighboring buildings and is easy to remember.
I followed this order: the reader's situation, why the place caught my attention, its features, then the official account and visitor information. That sequence can serve the next edition, but the place details, photos, and reason to feature it must be checked anew.
Even when I delegate a draft, I recheck opening hours, prices, and account details before publication. I also kept a rule against writing as if I had visited when I had not. The format can repeat; last week's fact-check cannot simply carry over.
Open the prompt for writing a caption after the cards
An example adapted from the actual work requests. Replace the files, account, and edition details with your own.
Approved card copy: [latest draft or file] Official sources and verification date: [links and dates] Voice reference: [an approved previous caption] Reader's situation: [when someone would need this post] Suggested next action: [one action, such as save, visit, or read more] Write an Instagram caption using only this material. Do not simply repeat the card sentences. Start with the reader's situation and explain why this subject deserves attention. Body order: A relatable opening; why it caught our attention; key features; verified information and sources the reader needs; a closing that helps them take the next step. Separate verified facts, the magazine's opinions, and unconfirmed details. Do not invent a visit or a recommendation I have not supplied. Do not invent prices, hours, or walking times without evidence. Check consistency with the final card title, sequence, and proper names. Separately list the verification date and anything to check before publication.
I could delegate repetition,
but deciding when it was done remained my job
Moving material and repeating exports felt easier. But this is not a workflow I can leave unattended from planning through publication. Here is how the requested work and my own checks were divided.
| Stage | Requested from AI | My decisions and checks |
|---|---|---|
| Planning | Organize existing material, research, and suggest titles and copy | Reader's situation, reason to feature it, and facts versus opinion |
| Design | Build editable templates and apply revisions | Photo selection, voice, type size, and layout |
| Video | Recomposite the latest title and logo, then export MP4 | Scene, legibility, cropping, and output-file checks |
| Finishing | Draft captions and organize files and review items | Check sources and current information; decide whether to publish |
AI's suggested photos did not always match the scene I wanted to show, and the writing's voice and emphasis needed work too. To avoid undoing my edits on the next request, I made sure AI would read the current version first and change only the requested parts each time.
This case does not measure production time or performance improvement. Connecting accounts, checking fonts, and correcting drafts also took time.
As the standards accumulate,
could I delegate more of the work?
Rather than automating everything at once, I want to delegate repetitive tasks one by one and record why I make changes. A request such as 'keep the copy and layout I revised' can become a rule to read the latest file first and leave approved parts unchanged for the next edition.
Set standards and continue the work
I had AI read the existing material, then reviewed the requested drafts, edits, and outputs as we progressed.
Make the next edition to the same standards
I plan to collect the reasons for revisions as production rules and check whether the same errors occur less often in the next edition.
Connect recurring runs with approval
I want reliable steps to run on a schedule, pause and notify me when information is missing or an error occurs, and proceed to publication only after review.
More rules do not make automation happen by themselves. They must work in the next edition, and exceptions need clear stopping points. I am still collecting production rules and working records; this is not a system with verified scheduled runs or automatic publishing.
For your own work,
start with the next piece of content
It does not have to be a magazine. Choose one recurring output, such as a weekly newsletter, product introduction, or teaching material. Rather than connecting every tool at once, I suggest starting in this order.
- Choose one task and define what done means.
For example, start by turning approved card copy into a caption draft. Define checks such as whether information is missing and whether the place sequence matches the cards.
- Provide a good example, current material, and the permitted scope of change.
Use one previous output and note the voice and structure to preserve. Separate this edition's approved copy from verified sources, and ask AI not to invent missing information.
- Record why you revised something, then test it in the next edition.
Write notes such as 'removed because it exaggerated' or 'held because there was no source.' Before expanding the scope, check whether the same mistakes recur and whether total working time, including revisions and review, actually falls.
Before your next request,
leave these six lines.
A suggested template for your own work, not a verbatim excerpt from the original production conversation.
This task: [content name and purpose] Latest file to continue from: [filename or accessible document link] Already approved: [copy, photos, and layout to preserve] What may change this time: [scope of revision] Why I revised it: [one sentence explaining the decision] Next task: [one task and its completion criteria] First review the latest file and these standards. Keep approved parts and revise only the requested scope. If material is inaccessible or standards conflict, tell me instead of guessing. When finished, summarize what changed and what still needs review.
Adapted from this project's requests so you can substitute your own work. These are not a finished automation program or an installable skill.
Questions before you begin
Is content being published automatically every week now?
No. In this case, AI carried out requested tasks and I reviewed them as production continued. Scheduled execution, error notifications, and a connection from approval to publication remain goals to test later.
Will these prompts alone let AI edit Figma?
A prompt does not create an integration. AI needs a connection and permissions to access and edit the document, and the required fonts must be available. If that connection is not ready, start with something you can verify in your current environment, such as organizing copy or drafting captions.
Do I have to move Illustrator work into Figma?
No. I used Figma so AI could continue from a canvas I had revised myself. What matters more than a specific tool is editable text and photos and a clear indication of which file is the latest version.
Will AI keep remembering material stored in a folder?
A folder stores material. You still need to specify which standards and latest files AI should read for the next task. Recording approved decisions and remaining work at the end gives a later request a starting point.
How do I decide whether a task is suitable for automation?
Start with work that has similar inputs and outputs each time and clear criteria for checking the result. If the direction changes substantially or evidence is missing, use AI for drafts and organization while keeping the decisions yourself.
Can I include client material or account information as it is?
Prepare only the material needed for the task and confirm what may be shared. Do not put passwords or API keys in drafts or prompts. A file being in a local folder does not mean a connected AI service will not process it.
Less repetition,
more attention to new stories
The magazine introduces new places, but making it involves many repetitive tasks. I began because I wanted to see how much of that repetition I could reduce. Deciding what to preserve and what to reconsider now feels as important as connecting the tools.
If revised copy, layouts, and the reasons for changing photos become standards for the next edition, perhaps the work that no longer needs a fresh explanation can gradually expand. I want to test that possibility again in the next project.
More than having AI make more,
I want to know whether I can repeat less
while keeping my standards.
When preparing your next piece of content, choose just one task you always repeat. Writing down the standards it needs before delegating it can be a small place to start.
Visit Itaewon Guide ↗